{"id":"arxiv/2606.29083","url":"https://arxiv.org/abs/2606.29083","tags":["math.NA","cs.LG","math.DS"],"score":0,"title":"Residual-Guided Dictionary Learning for Spectrally Accurate Koopman Approximation","domain":"math.NA","source":"arxiv","authors":["George Coote","Matthew J. Colbrook"],"reviews":0,"subject":"math.NA","summary":"Koopman theory promises linear structure in nonlinear dynamics, but numerical Koopman spectra are easy to compute and hard to trust. A finite EDMD matrix always has eigenvalues; the problem is that many of them may have nothing to do with the infinite-dimensional operator. In this paper we make spectral reliability the objective of dictionary learning. We train neural-network dictionaries not merely to predict the next snapshot, but to minimize Residual Dynamic Mode Decomposition residuals: operator-level a posteriori errors that test whether computed eigenvalues and modes are genuine…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-27","reactions":{}}